IMPROVING THE FUNCTIONING RELIABILITY OF THE INFORMATION MANAGEMENT SYSTEM ELEMENTS, USING BUILT-IN DIAGNOSTIC TOOLS

.


ABBREVIATIONS
IMS is a Information Management System; BC is a Battery Charge; FDD is a Fault Detection and Diagnostics; UC is a Universal Controller; CNN is a Convolutional Neural Network; MLP is a Multilayer Perceptron; TDT is a Test Diagnostic Tools; OD is a Object of diagnostic; DA is a Diagnostic agent; TD is a Test Diagnostics; FC is a Functional Control.

NOMENCLATURE
N cr is a number of critical elements of the system; N mCr is a number of middle-critical elements of the system; N nCr is a number of non-critical elements of the system P tr is a threshold power level; ∆P t is a power cost of one diagnostic test; ∆t is a diagnostic interval; t is a working time; i is a serial number of the element; P i (t) is a charge power of the i-element at time t, normalized to the range [0; 1]; CPU i (t) is a CPU load of the i-element at time t, normalized to the range [0; 1]; IMP CHAR is a numeric importance of the element characteristic; N i checks (t) is a distributed number of checks for ielement; N OE (t) is a number of operable elements at time t; N OE i (t) is a function to determine is i-element is operable;

INTRODUCTION
Nowadays, IMS are widely used. They allow corporations to automate their technological and business processes [1]. Current trends of their effective usage involve the use of wireless information transfer technologies to ensure the autonomy of the system and its elements. Autonomy is to increase the up-time of the IMS elements without additional supply of energy.
Wireless technologies and autonomy of the elements create additional factors of the system usage. These factors should be taken into account for planning routine maintenance. Because cost of the routine maintenance is directly proportional to the number of failures of the system elements.
For example, some fuel company has fully automated oil rig with autonomous fire-detection sensors. So, oil rig is commonly far away from command centre. If battery of the element (sensor) running out 2 times a week (or other failures), instead of one time -cost of the maintenance becomes 2 times more then it could be. Common cost includes the transfer cost, additional worker salary, etc.
So, improving reliability of the autonomous IMS elements should be considered for everyone who wants to have highly effective system and spend money efficiently.
The object of study was autonomous elements of information management system with wireless communication channels.
The subject of the study is development of a method for increasing the reliability of the functioning of autonomous elements of information control systems, including the creation of a model of an information control system and an algorithm for justified redistribution of diagnostic functions, as well as a software implementation of the developed algorithm.
The purpose of the work is increasing reliability of the autonomous elements of the wireless informationmanagement systems by increasing the operating time without recharging. by increasing the operating time of autonomous elements without recharging.

PROBLEM STATEMENT
It was found, that in an IMS system with built-in diagnostic tools, the existing diagnostic load balancing algorithms are unreasonable in relation to the current state of the system elements and thus negatively affect their reliability.
To solve the problem of unreasonable distribution of the diagnostic load, it was decided to develop a method for increasing the reliability of IMS elements and a software package that allows one to obtain a comparative characteristic of the developed and existing algorithms. Therefore, the main goal of the algorithm is to define the function (1) to recalculate (redistribute) the number of checks for each element before every diagnostic cycle. Input parameters: the classification of the elements of the system and their quantity (N Cr , N mCr , N nCr ), the threshold power level of the element (P tr ), cost for diagnostic test (∆P t ), the diagnostic interval (∆t). Current values of characteristics of the elements (in current work we use P i (t) and CPU i (t), but number of characteristics is expandable) and their weight coefficients (IMP CHAR ) are also should be used: A characteristic that will be an indicator of the effectiveness of the developed algorithm is an integral indicator of the number of operable elements of the system. To calculate it, we should declare power level of each element at time t and number of operable elements of the system at time t: Then the goal of the developed method will be to increase the numerical value of the resulting function over a finite time interval, the beginning of which coincides with the beginning of the functioning of the built-in diagnostic system: In addition, the criterion for the effectiveness of the developed algorithm will be a decrease in the average rate of change in the values of the resulting function (a smooth change in the number of workable elements). This allows reducing the cost of the batteries replacement.
It is also an important task to determine the conditions and boundary values of the parameters of the function, at which the developed technique is superior to the existing algorithms.

REVIEW OF THE LITERATURE
In this paper, reliability of the IMS elements is understood as up-time of the element, when the BC is higher than the minimum level. Minimum level of the elements BC should be defined as system requirement.
So, the IMS system reliability -is number of elements, whose BC is higher than minimum level.
The load of control processors significantly affects the discharge speed. In this regard, ensuring the reliability of the functioning of the IMS elements [2] by increasing uptime of the element is arises as a question.
One of the main tools to improve the reliability and fault tolerance of IMS elements -is usage of a progressive diagnostic system that meets modern requirements [3]. Diagnostic systems are divided by type into external and built-in. Each type is used in accordance with the reliability requirements of the system. The builtin test diagnostic system is characterized by the usage of diagnostic tools, built into the IMS elements. The advantages of the integrated [4] diagnostic systems includes, for example, the following: 1. The absence of the usage of the expensive and complicated in the deployment and administration diagnostic system.
2. More complete and adequate diagnostic information, characterizing a specific element of the IMS.
In this paper, the combined test diagnosis system will be investigated. UC is responsible for the distribution of diagnostic functions and functional control. Special tools, built into IMS elements structure, perform test diagnostics.
The use of built-in diagnostic tools [5] takes part of the productivity of control modules (processors, controllers) of IMS elements. The performance resources of the elements (testers), assigned to perform the diagnostic functions, are spent on the formation of the test, transmission and reception of data and their subsequent processing. In order to avoid the rapid discharge of batteries and malfunctioning, the article proposes a methodology for redistributing diagnostic functions between system elements. It allows reducing the load on the most important elements of the system or elements with a low battery charge. This will allow them to reduce the rate of discharge of the battery [6] and increase battery life without recharging. As a result, the reliability of the functioning of the IMS elements increases [7].
So, the problem noted above is quite relevant. At various times, the staff of IPU RAN named after V. A. Trapeznikova: Vedeshenkov V. A.
In particular, in [8] a method was proposed (Barsi-Grandoni-Maestrini model) for self-diagnosis of modules and communication lines of digital systems with reconfiguration. This allows stable multiple failures of a limited number of components. The method [8] is based on the principle of expanding domains.
In [9], existing methods of self-diagnosis were considered. Possibility of their application in the diagnosing a distributed radio-technical surveillance system framework was described. An approach to reconfiguring the system depending on the diagnostic results is proposed.
[10] discusses an approach to the design of FDD systems that have the potential to improve energy efficiency while reducing maintenance costs. To achieve this goal, virtual sensors and failure impact models are implemented that require only inexpensive measurements to detect the cause of failure, severely diagnose the failure, and evaluate performance degradation.
In [11], an adaptive built-in diagnostic system is considered. It actively uses transfer learning by developing an integrated approach for troubleshooting with different types of components. Two deep learning techniques are used: CNN and MLP, to train multiple basic models with a set of input data.
In [12], methods for detecting and diagnosing malfunctions used in the field of control in the aerospace, automotive and industrial fields are described. Each of these sectors uses specific techniques to meet different diagnostic needs. One of the most important gaps is the problem of resource efficiency in wireless information management systems.
Thus, it is worth noting that most of the papers on this topic consider diagnostic errors and problems in the data transmission channel as a reason for reconfiguring the diagnostic system. This is undoubtedly very important, but it is also important to take into account the current workload and performance of the system and its elements, because the diagnostic processes additionally load the resources of the system elements, leading to an increase in the likelihood of their wear and tear, failure and lower productivity. Figure 1 shows a block diagram of the developed TDT [13].

MATERIALS AND METHODS
This system contains functional elements (devices that directly perform the functions of the system), the reliability of which must be increased. A control and configuration device -UC is necessary for the distribution of functional tasks, collection results of execution and configuration of system elements. The functional elements communicate with each other and the UC via wireless information transmission channels [14]. Next, we will describe the built-in test diagnostics system.
In this system, it is assumed that each functional element of the IMS system, depending on the configuration, can be a test object or a tester. The configuration of each element is set by the UC, which Figure 1 -Element structure of the developed test diagnostics system also performs functional control (monitoring). The OD implements DA -a software and hardware module that performs the following functions: 1. Tester -formation of tests, transmission, reception and processing of diagnostic information.
2. TD -acceptance of the test and the formation of the response.
3. FC -receiving requests and generating a response about its technical condition (monitoring).
Thus, having entrusted the tasks of dynamic reconfiguration of the built-in test diagnostics system to the UC (without spending the resources of the functional elements (objects of diagnostics)), we will try to increase the reliability of the IMS elements.
In the developed methodology, the diagnostic model of the Preparate -Metz -Chen [15] was taken as the basis, in which the model is represented in the form of a graph whose vertices are the elements of the system and the edges are the inter-element diagnostic connections.
This means that the edge indicates the direction of the diagnostic tests. Also, in this model, each element is diagnosed once, by one diagnostic element -an agent. The novelty of the developed method consists in taking into account the performance characteristics of the IMS elements when distributing diagnostic functions between these elements. Such a system consists of elements of the same type, capable of diagnosing each other with the same tests in terms of resources and time spent.
You can divide the system into several blocks of the same type. In this case, the presented model will describe each block of the system separately, but collectively present the same diagnostic results as if it were the same for different types of elements.
To improve efficiency, you need to use the elements in different ways. The number of classes can be different (number of characteristics, number of elements). Features of the developed model 1. All elements must be divided into 3 classes: a. Critical -those elements that are entrusted with the most important and costly operations in the system, they must be diagnosed in the first place, and not used as a diagnostic tool.
b. Medium criticality -elements that are both diagnosed and are diagnosed. Half of this set of elements diagnoses at a certain point in time, when the remaining is diagnosed.
c. Non critical -they are constantly used as diagnostic elements. Least loaded system elements.
2. The UC covers the energy and computational costs of the reconfiguration of diagnostics. That is why, this process does not consume the power of diagnostic and diagnosed elements.
3. Input: ∆t, N cr , N mCr , N nCr , P tr . 4. At the first stage of the development of the algorithm, we will take ΔP t as a constant, but in a real situation -the value is a variable that depends on other parameters. On the Fig. 2 we see an example of reconfiguration of the purpose of diagnostic functions of a system consisting of 2 critical elements, 2 elements of medium criticality and 2 non critical elements. At the moment t, the numerical value of the performance of each element (the calculation is performed by the UC, it collects statistics before each cycle of test diagnostics) made it possible to determine the next diagnostic graph for the system (left graph in Fig. 1). It shows that it is imperative to diagnose critical elements, and at time t with minimal costs, elements 1, 5, 6 will cope with this. At the next time moment t + Δt, the performance indicators of the elements changed, as well as the numerical estimate, which allowed the UC to redistribute the diagnostic functions.
Let's present an analytical model of the built-in diagnostic system reconfiguration algorithm.
To develop a reconfiguration algorithm, we introduce a model for quantitative assessment of the parameters of the method: 1. N checks . This value characterize the number of needed checks in this time interval: 2. The maximum number of checks for which the resources of each element in the sum are enough: 3. Now we introduce the performance characteristics of each element. We use the dependence from only one characteristic, the expansion will be performed later in this article: 4. Number of checks for each element: where N checks (t) will be equal (4) or (5) depending on the conditions described below. Thus, all the quantities necessary for the development of the reconfiguration algorithm are determined.
Due to the development of wireless technologies [2], autonomous IMS elements have become widespread. Examples of autonomous IMS elements can be (autonomous TP-Link TL-MR3040 routers), energy autonomous products from CISCO (Cisco Aironet 1240), autonomous sensors for monitoring leaks and fire safety in oil fields (Lukoil). It is also relevant for mobile devices and AD-HOC networks [16], since they certainly have an autonomous power supply of the elements [17].
For various information and control systems, the performance characteristics of its elements may be different [18]. Thus, one of the requirements for the developed algorithm is scalability in number of characteristics, which affect the performance of elements.
As you know, a decrease in the battery charge level leads to a malfunction of the elements, which leads to a decrease in such a reliability indicator as mean time between failures. Most modern cells have built-in diagnostics that take up part of the performance of processor devices and, accordingly, accelerate battery discharge. Thus, with redistributing the load of elements with diagnostic functions, it is possible to selectively reduce the discharge rate and increase the reliability indicators of the IMS elements.
A new algorithm for reconfiguring the built-in subsystem of test diagnostics has been developed to perform the main functions of the UC. As a first approximation, we consider one performance characteristic and take ΔP t = const.
Algorithm diagram is shown on the Fig. 3. 1. The first step of the reconfiguration algorithm is to calculate the number of checks that will fall on each diagnostic element at time t with an interval of Δt. Let us calculate the total number of checks for each moment of time t according to the (4).
2. The second step is to determine the possible number of checks at a given time, depending on the current battery charge. This number does not depend on the number of characteristics that must be taken into account in the distribution of diagnostic functions. We make the calculation using the (5). If, N i (t) ≤0, then the element is removed from the category of diagnostics with the corresponding notification of AST. Also, if: In this case, a notification is sent that not all necessary checks will be performed at the current time. The necessary action in this case is to replace the batteries. And if: then a notification is sent that the current check is the last full check (that is, all tests will be carried out) and the next one will be incomplete.
So, if two first conditions are false, then:  3. The next step is to distribute the diagnostic load depending on the performance characteristics of the cells, and in this case -the current battery charge. For this, we introduce the required coefficient according to the formula (6). Thus, the number of checks per element is calculated by the formula (7). 4. The last step of the algorithm is to round the number of checks to an integer in such a way that the number of checks for an element with a large current charge is rounded up, and for a smaller one, accordingly, down.
5. Carrying out diagnostics according to the performed reconfiguration, sending the results to the UС. Waiting for the end of the diagnostic interval. Go to step 1.
The input data of the algorithm are set at the stage of creating a system based on the class of tasks solved by the system and the necessary requirements for the reliability of the system. Now, the method of the expansion of the number of considered characteristics will be presented.
In step 3 of the algorithm, the distribution coefficient of checks by the current level of battery charge was presented. To take into account additional parameters, a complex coefficient is required. This coefficient is proposed to be calculated according to the following algorithm.
Let us add one more characteristic of the IMS element -the processor load at the current time. High processor utilization leads to decreased performance, slower performance and faster battery drain. This dependence is because with a high processor load, the diagnostic test execution time will increase, which will undoubtedly lead to an increase in battery consumption. Therefore, a reasonable distribution of diagnostic tasks between various elements can increase the battery life of an element, as well as improve performance and reliability indicators.
An example of a common specification is processor frequency. As a particular characteristic, one can suggest, for example, the distance of signal transmission between the elements of the system. Here is an example of a modified calculation of clause 3 of the algorithm: The difference is necessary to take into account the requirements for improving the characteristics, because the more the processor is loaded, then the worse the conditions for this element to diagnose. For 2 characteristics under consideration, the complex coefficient will looks like: To numerically determine the importance of characteristics, several approaches can be used. For example: 1. Mathematical (average, weighted average).
2. The theory of the importance of criteria by Podinovsky [19] 3. Coefficient of concordance (consistency of expert judgment) 4. Fuzzy logic. For this algorithm, it is advisable to use the average approach to determine the weight coefficients. Its advantages are ease of use, clarity of the distribution of importance characteristics, a large number of algorithms in which it is already used [20]. If necessary, the method can be changed.
Development of a software model in which the developed diagnostic load distribution algorithm is implemented.
In order to check the correctness of the developed algorithm, a software was developed that simulates the operation of the developed algorithm and shows a comparison with other algorithms, when the entire diagnostic load is assigned to one element until its failure. Let us call this algorithm "1 → all". There are two types: simple and modified. Simple's gist is, briefly, that one item is assigned as a tester until it fails. Then the next element takes over its responsibilities and works until it fails. Modified algorithm reassign the tester before each diagnostic cycle. Program was written in Javascript (for cross-platform) using the EmberJS framework [21].
Below are the results of the algorithm. When entering input parameters, the program generates the entered number of elements, assigning them random initial values of the battery charge and processor load. From this starting point, the algorithms being compared operate. An example of the generated initial state is shown in Fig. 4.
In the figure, on the left side, we see a panel for entering the system parameters, necessary for the developed algorithm to work. On the right, we see the visualization of the generated elements of the information system according to the entered parameters. Inside each element there are 2 columns -the left one is responsible for the charge level of the element, the right one is for the current CPU load. Before each diagnostic cycle, the CPU value is randomly generated again, the battery power value is reduced depending on the number of diagnostic tests performed. Functions to calculate the needed values (2,3) to estimate the effectiveness of the developed algorithm are presented in program, so: The result of the program's work (Fig. 5) is a graph that demonstrates the calculated by formula (8) number of operable elements. The OE characteristic will be calculated like: 0 ( ) .

EXPERIMENTS
We will take 3 system types with different number of elements: 20, 100, 500. It means that the first system will have 20 critical, 20 middle-critical and 20 non-critical elements, second will have 100 critical, 100 middlecritical and 100 non-critical elements, etc. And we will take 2 different ∆P values to see, how the complexity of the test (resource needed) affects algorithms efficiency.
Let us analyze graphs. We must visually pay attention to the value of the area under the graph (OE) and smoothness in function value decrease. Numerically OE can be calculated by (9), but here we can visually analyze this value as area under the graph. On Fig. 6 and Fig. 9 we can see, that in system with low number of elements, developed method works less efficiently then the modified "1 → all" method. However, at the same time we can see that it has the smoothest decline between all methods. It means that elements battery discharge is flowing. This type of discharge is the best way to extend battery and increase the element reliability.
We can see, that increasing the number of elements leads developed algorithm to work more efficient. Number of operable elements on Fig. 7, Fig. 8, Fig. 10, Fig. 11 is more than in other methods. So we can conclude, that if systems have a lot of elements, it's better to use developed algorithm.
At the same time, we can see the intersection of graphs and method "1 → all" becomes more efficient, when batteries charge is low. We can conclude that in further researches the combined algorithm can be invented. We can take the best from both algorithms and create an algorithm, which will choose the most effective method in current time and situation.
The efficiency become a little bit higher with the ∆P cost increasing. We can see, that the difference between operable elements on Fig. 8 is less than in Fig. 11. It means that the algorithm works well in highly loaded systems.

RESULTS
As a result of the theoretical studies, an IMS model has been developed, to improve the reliability indicators of which it is proposed to use an integrated test diagnostics system. Methods to define the criteria weight were presented. One of them was chosen in the developed model and reasons of it are also presented. A mathematical model has been created for calculating the main indicators of the functioning of the IMS elements. An algorithm for redistribution of the diagnostic load has been developed, which uses the results of calculations according to the proposed model and allows increasing the performance indicators of the system elements.
A software model has been developed that implements the developed algorithm and allows it to be compared with existing algorithms. A study of a software model with various parameters was carried out and, based on the results of software modeling conclusions were drawn about the possibilities of improving the algorithm, and directions for further scientific research were formulated.

DISCUSSION
The several existing algorithms where chosen for the experiments. Experiments showed some advantages and disadvantages of the developed algorithms as well as advantages and disadvantages of the other algorithms. It helps to understand the best conditions for algorithm usage and improvement possibilities. Combination of best results can help to update the developed algorithm to make it more efficient in other conditions. Therefore, the best conditions to use the developed algorithm are when the number of elements is higher than 100 (of each type). With an increase in number of elements, we can see an increase in number of operable elements. In system with 100 elements of each type the average increase in the number of work items is 6% compared to modified «1 → all» algorithm and 35% compared to «1 → all» . In system with 500 elements -7% and 45% resp. Also, with the increase in test power cost, the developed algorithm increases the number of operable elements by 2%, meanwhile other algorithms don't show the same result. It shows, that algorithm also works well in systems with high loaded elements.
These results can help to increase the IMS system elements reliability. Smooth and slow discharge of the elements battery and CPU level as a reconfiguration reason can help to extend the life of the battery and functional parts of the elements, which also saves money for routine maintenance.

CONCLUSIONS
This article presents a developed methodology for increasing the reliability of IMS elements using built-in diagnostic tools. The characteristics of the IMS elements (processor load and battery discharge rate) were analyzed, which significantly affect the reliability of their operation. It is shown that the means of test diagnostics tools (TDT) also affect the indicated parameters, but, to reduce their negative influence, a technique for their reconfiguration is proposed in the work. It lies on reasonable reconfiguration of the objects of the built-in test diagnostics subsystem based on taking into account the performance factors (processor load) and energy efficiency (battery discharge rate). The developed methodology is scalable in number of characteristics.
The scientific novelty of the results is: developed diagnostic model of the built-in diagnostic system and algorithm for its reasonable reconfiguration. Also, to confirm theoretical assumptions, a software simulation program was developed. It makes possible to qualitatively and numerically evaluate the advantages of the developed algorithm.
The practical significance of the results of the work lies in the improving the reliability of autonomous elements of the IMS. Increasing the number of operable elements in the current moment of time will help to reduce costs of the element's service and usage.
Prospects for further research are expected in the development of algorithmic and software tools for full simulating. It's also expected to improve the method of calculation of the ∆P value. It's important, because its value is not constant, it's value should be calculated as two variable function, depends on current processor load and current charge level of the element. Also, it's interesting to combine the best results of several algorithms and update the developed one.